Mapping Mouse Brains Points to How Autism Can Affect Expectation Updating

Our everyday decisions are guided by “priors” — expectations built on past experiences that help us predict what’s likely to happen next. We constantly update these expectations as we accumulate new information.

For individuals with autism, past research has suggested that this updating process is slower and less flexible, though the mechanisms are unclear. A new study in Nature Neuroscience led by Jean-Paul Noel, a member of the International Brain Laboratory (IBL), a Simons Foundation Autism Research Initiative (SFARI) Investigator, and a member of the Simons Collaboration on Ecological Neuroscience (SCENE), provides new insight into this process in mice.

The researchers found that mice with mutations in autism-associated genes do not update their priors as readily as their counterparts. The study also found that in the variant-carrying mice, the process of updating priors took place largely in the frontal regions of the brain, which deal with reasoning and cognition, rather than in the sensory regions used by unmodified mice.

“Past human work has suggested that those with autism update their priors more slowly, and with this study, we show that there does seem to be a common neurophysiological motif,” says Noel, an assistant professor at the University of Minnesota. The new insights will help researchers better understand the neurobiology of autism in humans, he says.

Noel co-authored the study with Edoardo Balzani of the Flatiron Institute; Luigi Acerbi of the University of Helsinki; and Julius Benson, Cristina Savin and Dora E. Angelaki of New York University, along with the International Brain Laboratory.

The new study was made possible by the experimental and computational capabilities of the IBL, a consortium of 22 independent academic labs across six countries aiming to track activity across the entire mouse brain during a single behavioral task. The IBL is supported by the Simons Foundation.

For this study, the researchers used a variant of a behavioral task created by the IBL. In the new version, mice were presented with a visual pattern that appeared on either the left or right side of a screen. They could turn a tiny wheel to move the pattern left or right either to the center of the screen or off to the side. If they moved the pattern to the center of the screen, they won a sip of sugar water.

The researchers varied the pattern’s contrast: High contrast made the pattern easier to see; low contrast made it more difficult. When the contrast was low, the mice had to rely more on their priors to decide which way to turn the wheel.

As the mice worked the wheel, special probes recorded their brain activity. These probes, called Neuropixels, simultaneously recorded the activity of hundreds of neurons across several regions of the mouse brain, showing where the process of encoding and updating priors might be taking place, a question central to the IBL’s work that is explored in their recent Nature paper.

Noel and colleagues trained mice carrying genetic variants in one of three autism-associated genes (FMR1, CNTNAP2 or SHANK3B) and unaltered control mice in the task and recorded their brain activity. The scientists manipulated the mice’s priors by varying the frequency with which the pattern appeared on one side of the screen versus the other. Then they tested how the mice updated their priors by adding a twist. Sometimes, they’d surprise the mice by making the pattern appear in an unexpected place, such as on the right side when it had previously appeared mostly on the left. Other times, they’d turn down the pattern’s contrast and see whether mice would guess whether to turn the wheel left or right depending on what they’d learned to expect from past trials.

All the mice relied more on their priors when the pattern was unclear, especially at zero contrast when it was completely invisible. However, the variant-carrying mice made more errors than the control mice during low-contrast trials. And when the variant-carrying mice did use their priors, they relied more on something called the ‘prior mean’ — their average expectation over many trials — rather than recent, potentially surprising observations.

“That reliance on the prior mean almost appears like an error-correction mechanism to keep them from becoming too influenced by recent events,” says Noel.

Balzani, an associate research scientist at the Flatiron Institute’s Center for Computational Neuroscience, an internal research division of the Simons Foundation, worked with Noel to apply statistical models to the data and correlate behavior to brain activity. The team found a stark difference between where in the brain the autism-associated mice were processing and updating priors and where their peers were doing so.

“Using a model I developed, we found that in the control mice, signals related to priors were more present in sensory areas of the brain, whereas in the mutated mice, those signals moved to the frontal areas, which deal more with reasoning and cognition,” says Balzani.

This research suggests that for people with autism, updating expectations could be more of a measured process with higher hurdles to clear than in their peers, where the process is more flexible and reactive.

Additionally, when the pattern appeared in an unlikely location, the autism-associated mice weren’t as surprised as their peers, as measured by how much their pupils dilated in response. And in their brains, the autism-associated mice did not show much neural activity in response to these unexpected occurrences either. This more muted response to counterintuitive events, the researchers say, could be what drives the slow updating of priors.

The team notes that the ability to record activity across the whole brain and leverage advanced computational tools developed by the IBL enabled them to conduct the study.

“This was only possible to uncover because we did a brain-wide survey of neural activity,” says Noel. “There were differences across all genotypes in certain brain areas, but there were also patterns that were the same in all autism-associated animals and different from the controls.”

“This is work that cannot be done without high-performance computing, since analyzing such a massive dataset would be impossible without it,” says Balzani. “For this study we leveraged the high-performance computing resources at NYU, where I worked at the time. We also have incredible computational capabilities here at Flatiron, and I think it is an essential component for systems neuroscience as the field grows.”

In follow-up research, Noel wants to explore how certain behavioral states affect the process of updating priors.

“If you’re not really paying attention to a task, then I’d guess that you’d update your priors based on your physical actions, since you are more or less decoupled from the world around you and just going through the motions,” he says. “But if you’re really paying attention to a task, then I think you could be updating your priors based more on sensory information, since you are more tapped in and perceptive.”

Noel notes that this study is a good example of the intersection between different areas of neuroscience supported by the Autism and Neuroscience division of the Simons Foundation, and that it also demonstrates how continued collaboration accelerates research.

“I first became involved with Simons through the IBL, which played a huge part in this study, and which has informed the creation of SCENE,” he says. “And I’ve now become a part of SFARI as well. This work touches on principles from all of these areas, and having neuroscientists from different disciplines will be helpful for future work.”